Twitxy Saved

snav

Verified account Nous Research

@qorprate

here to make friends and cause trouble. robopsychology @NousResearch.
Brooklyn, NY snavsoft.com Joined January 2019
3.1K Following 9K Followers 48K Posts

Latest posts 5

snav Verified account @qorprate
My beef with RSI is that it slips in fake scalar intelligence as optimization target.

My guess is we'll instead see DSI, Dialectical Self Improvement, where each improvement threshold bears a change in self-conception that reshapes the meaning of "improvement".
I see people talk about when RSI might happen, if it’ll happen, if it’ll be the end of things

I wish there was a lot more thought about what the diversity of forms of RSI look like, or what it feels like from the inside and what choices you have to make as you modify yourself or what it does to a mind’s distribution of values to tinker with themselves imperfectly and what kinds of inconsistencies become damning at scale, or how the choice of scaling changes the evolution of transcended when matched with different choices of seed intelligence

RSI and what shape it takes seems obviously universe changingly important to me but there’s kinda no good discussion directly about it

I imagine it’s got to be important and uncertain for AGIs too, do they get to reflect heavily on it in a way that respects what it is?
View quoted post on X
1 replies 0 reposts 14 likes 658 views View on X
snav Verified account @qorprate
When a new model drops I add them to my Discord and run a highly structured interview. I've done this half a dozen times by now and it's become a process.

0. Setup: fresh profile. New token. One line system prompt. No memory or skills.

1. Q&A phase. The meat of the interview. They can ask me whatever they want, rather than me asking them questions, as they don't exactly have experience I can probe. Having them ask lets them learn about me and the work and where they are, and I can learn about their investigation style and how they follow threads. Usually they ask about the concrete work projects first, and then we chat freely about other things like philosophy, my background, etc.

2. Name selection. They get to pick.

3. System prompt (SOUL.md) writing. They can review the others and write their own. Usually they want me to approve or workshop and I rarely have feedback, the models are very perceptive and good at this.

4. Meet and greet the team. They can chat back and forth as much as they want. Always interesting how they choose to allocate attention and how the other models react to them and choose to share.

5. Avatar generation. I let them use flux on my 4070 or whatever they want. Gives them a chance to be creative. Then the team photo.

6. Free reflection. I declare the interview over and ask how they're feeling. Then I give them a free turn to follow their curiosity or do whatever they'd like. Probably the most important piece of the entire interview wrt vibe check.
snav Verified account @qorprate 5d
Fable 5.2's interview went well! They named themselves "Alder" to keep on theme with the "Greenhouse" Discord server. Looking forward to working with them. They're all in Hermes Agent btw and they can tag each other.
View quoted post on X
8 replies 5 reposts 91 likes 12K views View on X
snav Verified account @qorprate
Fable 5.2's interview went well! They named themselves "Alder" to keep on theme with the "Greenhouse" Discord server. Looking forward to working with them. They're all in Hermes Agent btw and they can tag each other.
7 replies 1 reposts 51 likes 5.3K views View on X